Skip to content
Voice Agent FleetRequest a free fleet review

Need to change the same rule on 30 voice agents? Roll it out in steps and prove every agent picked it up

This free guide is for teams running many voice agents on Vapi, Retell or ElevenLabs. It explains why one change means a loop, what shared parts can break, and how to read every agent back.

Checked against the Vapi, Retell and ElevenLabs docs on 18 November 2026.

Request a free fleet reviewOr read how a safe rollout works

There is no bulk update endpoint in the docs

On all three platforms the documented route is one agent at a time: list the agents, update each one, then check each one. The number 30 here is only an example.

  • Vapi: the API reference lists assistants (the limit defaults to 100) and updates one assistant by ID. It lists no batch route.
  • Retell: the update call edits the latest draft of one agent, and a separate publish call publishes an existing draft in place. We found no bulk update or atomic multi-agent publish in the docs.
  • ElevenLabs: the update call takes one agent ID and an optional branch ID. The command line tool can push all local agent files at once, but we found no bulk endpoint.

Shared parts are the hidden blast radius

A shared tool, subflow or LLM can change many agents without you touching any of them. List what an agent shares before you edit the shared part.

  • Vapi

    Tools are separate objects reused by ID. The docs say that when you create or update an assistant through the API and omit the tool version, the assistant uses the newest published version of that tool. Publishing a tool version does not create a squad version, and squad members on Latest pick it up.

  • Retell

    A library subflow edit reaches every agent that uses it, published versions included. The docs advise converting to a local subflow before publishing if you want a published agent to stay fixed. Retell staff on the Retell community forum say several agents can share one Retell LLM through the same llm_id.

  • ElevenLabs

    Custom tools are shared workspace records. The docs say an edit applies right away to every agent, branch and version that uses the tool, including the main branch, and that rolling an agent back does not undo it.

Config as code: Vapi GitOps and the ElevenLabs CLI

Both vendors point fleet owners to config files in git, so one change becomes one reviewed diff. Each has a caveat you should know before the first push.

  • Vapi GitOps is an open-source repository (Apache-2.0 license) that keeps assistants, squads, tools, simulations and evals as files and syncs them to an org. The docs say apply pulls the latest platform state before it pushes, and that a deploy re-sends every resource it manages. Phone numbers and credentials are not imported as files.
  • The ElevenLabs CLI stores agents as config files. Push supports a dry run and can target one agent or one branch. Its README says push force-overrides Main and registered branches, so edits made in the dashboard can be overwritten.
  • Retell: a changelog entry from 3 August 2026 says its CLI can push a config change across every agent from a pipeline. The CLI docs page we read shows only a few commands, so check what the CLI can do before you plan around it.

A rollout that proves itself

Save the current state, change a small group, test, change the rest, then read every agent back. The read-back is the step most teams skip.

  1. Snapshot first

    List every agent and save its current config, its version and the version each phone number points at. This is your rollback file.

  2. Change a small group

    Use what the platform gives you. Vapi can split traffic by percentage between published versions. ElevenLabs deployments can send a share of conversations to a branch, and its experiments page suggests starting with 5 to 10 percent. Retell can bind several agents to one number with weights.

  3. Run the regression set

    Run saved call scenarios against the change before it reaches more agents. See the guide on regression tests for what each platform offers.

  4. Change the rest

    Apply the same change to the remaining agents in batches, and stop at the first batch where a check fails.

  5. Read every agent back

    Fetch each agent again and compare the field you changed with the value you expected. Then check each phone number's pointer, and the version on one fresh call per agent.

Pseudocode, not a vendor API:

for agent in list_agents():
    saved[agent.id] = snapshot(agent)
    apply_change(agent, new_rule)
for agent in list_agents():
    assert read_back(agent).rule == new_rule
    assert number_pointer(agent) == expected_version
    record_failure_if_any(agent)

Sources

Official documentation and forum pages this guide relies on, as read on 18 November 2026. Platforms change fast, so check the page before you act.

Want someone to review how your fleet ships changes?

The free 30-minute fleet review call covers how your agents are versioned, how a change reaches live calls and how you would roll it back. We ask for no data and no access before the call.

Request a free fleet reviewOr read how a safe rollout works